8
$\begingroup$

Let $B$ be a paracompact space with the property that any (topological) vector bundle $E \to B$ is trivial. What are some non-trivial examples of such spaces, and are there any interesting properties that characterize them?

For simple known examples we of course have contractible spaces, as well as the 3-sphere $S^3$. This one follows from the fact that its rank $n$ vector bundles are classified by $\pi_3 (BO(n)) = \pi_2 (O(n)) = 0$. I'm primarily interested in the case where $B$ is a closed manifold. Do we know any other such examples?

There is this nice answer to a MSE question which talks about using the Whitehead tower of the appropriate classifying space to determine whether a bundle is trivial or not. This seems like a nice tool (of which I am not familiar with) to approaching this problem. As a secondary question, could I ask for some insight/references to this approach?

EDIT Now that we know from the answers all the examples for closed $3$-manifolds, I guess I can now update the question to the case of higher odd dimensions. Does there exist a higher dimensional example?

$\endgroup$
5
  • 2
    $\begingroup$ The main idea is to classify the maps $X\to Y$ up to homotopy, for two "good" topological spaces, which is obstruction theory. If you are not familiar with Whitehead towers, you could instead pick a cellular structure on $X$ (if it exists) and extend maps by skeleton. This is described in Steenrod's book The Topology of Fibre Bundles or Fomenko–Fuchs' textbook. $\endgroup$
    – Z. M
    Commented Feb 25, 2022 at 19:11
  • 1
    $\begingroup$ Another good source for obstruction theory is Whitehead's Elements of Homotopy Theory. $\endgroup$ Commented Feb 26, 2022 at 2:24
  • $\begingroup$ @Z.M I see, thanks for the references. I'm wondering in particular about the use of the if and only if in Qiaochu's answer, since usually a lack of obstructions does not imply existence, in many cases. In this context, do we truly get that his map $f$ near the beginning is nullhomotopic iff all the appropriate classes vanish? $\endgroup$ Commented Feb 26, 2022 at 16:37
  • 1
    $\begingroup$ The OP recently crossposted the question to MSE. Jason DeVito gave an excellent answer there. $\endgroup$ Commented Mar 22, 2022 at 5:03
  • 1
    $\begingroup$ @PaulCusson: It depends on how powerful your obstruction theory is. Often the complete obstructions can be subtle to compute, but there are some very nice situations where the complete obstruction agrees with more computable reduced obstructions. For example, the Whitney obstructions are a type of complete obstruction, and they have mod-2 reductions that are the Stiefel-Whitney classes, which are an order of magnitude more computable. That said, sometimes the a Whitney obstruction agrees with the Stiefel-Whitney class. $\endgroup$ Commented Mar 23, 2022 at 17:34

3 Answers 3

7
$\begingroup$

Let $B$ be a closed manifold with such that every vector bundle is trivial. Then $H^1(B; \mathbb{Z}_2) = 0$, otherwise there would be a non-trivial line bundle. Therefore every bundle over $B$ is orientable and $B$ itself is orientable. Orientable rank two bundles over $B$ are classified by $H^2(B; \mathbb{Z})$, so we must have $H^2(B; \mathbb{Z}) = 0$. It follows these conditions that there are no examples with $\dim B = 1, 2$. If $\dim B > 2$ we then have to consider the possibility of non-trivial bundles of rank at least three.

Suppose now that $\dim B = 3$. As $B$ is closed and orientable we have $H_1(B; \mathbb{Z}) \cong H^2(B; \mathbb{Z}) = 0$ by Poincaré duality and $H^3(B; \mathbb{Z}) \cong \mathbb{Z}$. It follows that $B$ is an integral homology sphere - note, the condition $H^1(B; \mathbb{Z}_2) = 0$ is superfluous in this case as $H^1(B; \mathbb{Z}) \cong \operatorname{Hom}(H_1(B; \mathbb{Z}), \mathbb{Z}_2)$ by the Universal Coefficient Theorem. We still have to consider the possibility of non-trivial bundles of rank at least three.

Suppose $E \to B$ has rank greater than three, then $E\cong E_0\oplus\varepsilon^k$ for some rank three bundle $E_0$, see this answer; in particular, we only need to consider the possibility of a non-trivial vector bundle of rank three. Suppose then that $\operatorname{rank} E = 3$. As $E$ is orientable, it has an Euler class $e(E)$ which is the first obstruction to a nowhere-zero section (it is also the only obstruction because $\operatorname{rank}E = \dim B$). As $E$ has odd rank, the Euler class of $E$ is two-torsion, but $e(E) \in H^3(B; \mathbb{Z}) \cong \mathbb{Z}$ which is torsion-free, so $e(E) = 0$. Therefore $E \cong E_0\oplus\varepsilon^1$ where $\operatorname{rank}E_0 = 2$. As we already know rank two bundles over $B$ are trivial, we see that $E$ is also trivial.

In conclusion, we have the following:

Let $B$ be a closed three-manifold. Every vector over $B$ is trivial if and only if $B$ is an integral homology sphere.

Note, we only considered real vector bundles above, but the same is true for complex vector bundles since every such bundle is the direct sum of a trivial bundle and a complex line bundle, but the latter are classified by $H^2(B; \mathbb{Z}) = 0$.

The fact that every vector bundle over a three-dimensional integral homology sphere is trivial can also be seen using Quillen's plus construction (also see section $\mathrm{IV}.1$ of Weibel's An Introduction to Algebraic K-Theory). As $B$ is an integral homology sphere, its fundamental group $\pi_1(B)$ is perfect. By the plus construction, there is a simply connected CW complex $B^+ = B^+_{\pi_1(B)}$ and a map $q : B \to B^+$ inducing isomorphisms on homology satisfying the following: if $f : B \to X$ is a map with $\ker f_* : \pi_1(B) \to \pi_1(X)$ equal to $\pi_1(B)$, then there is a map $g : B^+ \to X$, unique up to homotopy, such that $f = g\circ q$. As $H^1(B; \mathbb{Z}_2) = 0$, every bundle over $B$ is orientable and hence classified by a map $f : B \to BSO(n)$. Since $\pi_1(BSO(n)) \cong \pi_0(SO(n)) = 0$, the kernel of $f_* : \pi_1(B) \to \pi_1(BSO(n))$ is $\pi_1(B)$ so $f = g\circ q$ for some map $g : B^+ \to BSO(n)$. Note that $B^+$ is a simply connected CW complex with $H_*(B^+) \cong H^+(S^3)$, so $B^+$ is homotopy equivalent to $S^3$ by the homological Whitehead Theorem. As $\pi_3(BSO(n)) \cong \pi_2(SO(n)) = 0$, the maps $g$ and $f$ are nullhomotopic, so $f$ classifies the trivial bundle. Replacing $BSO(n)$ with $BU(n)$ yields the same result for complex vector bundles.

As for higher-dimensional examples, note that they must have odd dimension. To see this, suppose $\dim B = 2m$. Choose a degree one map $\varphi : B \to S^{2m}$ and consider the bundle $\varphi^*TS^{2m} \to B$. As $e(TS^{2m}) \neq 0$ and $\varphi^* : H^{2m}(S^{2m}; \mathbb{Z}) \to H^{2m}(B; \mathbb{Z})$ is an isomorphism, we see that $e(\varphi^*TS^{2m}) = \varphi^*e(TS^{2m}) \neq 0$ and hence $\varphi^*TS^{2m}$ is non-trivial. If one considers complex bundles instead, the same argument works by replacing $TS^{2m}$ with a complex vector bundle $E$ with $e(E) = c_n(E) \neq 0$ - such a bundle always exists as $\operatorname{ch} : K(S^{2m})\otimes\mathbb{Q}\to H^{\text{even}}(S^{2m}; \mathbb{Q})$ is an isomorphism.

One last comment about dimension five. As was established by Jason DeVito here, $B$ must be a rational homology sphere. If a five-dimensional example exists, I claim it is also a $\mathbb{Z}_2$ homology sphere. To see this, first note that we have $H^1(B; \mathbb{Z}_2) = 0$ from above. Now, if $v \in H^2(B; \mathbb{Z}_2)$, then there is a bundle $E \to B$ with $w_2(E) = v$ if and only if $\beta(v^2) \in H^5(B; \mathbb{Z})$ is zero where $\beta$ denotes the mod $2$ Bockstein, see this answer. As $\beta(v^2)$ is two-torsion and $H^5(B; \mathbb{Z}) \cong \mathbb{Z}$ is torsion-free, we see that $\beta(v^2) = 0$ for every $v \in H^2(B; \mathbb{Z}_2)$, so every such class arises as $w_2(E)$ for some $E$. Therefore, we must have $H^2(B; \mathbb{Z}_2) = 0$. The claim now follows by Poincaré duality.

$\endgroup$
6
$\begingroup$

Here is one constraint, which seems relevant in light of Michael Albanese's answer:

Claim: Let $B$ be a closed orientable odd-dimensional manifold with no stably nontrivial complex vector bundles. Then $B$ is a rational homology sphere (of odd dimension).

Proof:

  1. By Bott periodicity, $\widetilde{KU}^\ast(B)$ is concentrated in odd dimensions.

  2. Now, $\widetilde H^\ast(B; \mathbb Q[\beta^\pm])$ (with $|\beta| = 2$) is the rationalization of $\widetilde{KU}^\ast(B)$, so it is also concentrated in odd dimensions. So $\widetilde H^\ast(B; \mathbb Q)$ is also concentrated in odd dimensions.

  3. Now since $B$ is odd-dimensional, it follows by Poincaré duality that $B$ is a rational homology sphere.

$\endgroup$
1
  • 1
    $\begingroup$ Its interesting to note that the real analogue of the claim is also true. See Jason DeVito's answer to the MSE crosspost. $\endgroup$ Commented Mar 22, 2022 at 5:01
4
$\begingroup$

Here is another obstruction.

Suppose $M^n$ is a closed simply connected manifold which admits only trivial vector bundles. Then $M$ cannot be a $\mathbb{Z}/2\mathbb{Z}$-homology sphere, unless $n=3$.

I'm not sure if the hypothesis that $M$ is simply connected is necessary, but it's certainly necessary in the proof.

Together with the end of Michael Albanese's answer, this implies that every simply connected $5$-manifold admits a non-trivial vector bundle.

Proof Sketch: By the work in the other answers, we already know that if $M$ admits only trivial vector bundles, then $M$ is odd dimensional, orientable, and it has the rational homology of a sphere. Thus, we may assume $n\geq 5$.

Now, assume for a contradiction that all the torsion in $H^\ast(M)$ is of odd order. Let $k$ denote the least common multiple of the orders of the torsion and note that $k$ is odd.

Because $M$ is orientable, it has a degree $1$ map $f:M\rightarrow S^n$. We will construct a non-trivial vector bundle on $M$ by pulling back a non-trivial bundle $E$ over $S^n$ along $f$. When $n\neq 7$, we may use the tangent bundle $E=TS^n$. When $n=7$, we let $E$ denote the rank $3$ vector bundle corresponding to a generator of $\pi_7(BO(3))\cong \pi_6(O(3)) = \pi_6(S^3) = \mathbb{Z}/12\mathbb{Z}$. Let $\phi:S^n\rightarrow BO(s)$ denote the classifying map of the bundle $E$ (where $s = n$ if $n\neq 7$, and $s= 3$ when $n = 7$.)

We claim that $f^\ast E$ is a non-trivial vector bundle over $M$. To see this, note that since $M$ is simply connected and all torsion is annihilated by $k$, there is a map $g:S^n\rightarrow E$ of degree $k^r$ for some integer $r\geq 1$. (See the answer here for a proof.)

It is enough to show that $g^\ast(f^\ast E)$ is a non-trivial vector bundle over $S^n$. Of course, it is enough to show that the map $\phi\circ g\circ f:S^n\rightarrow BO(s)$ is homotopically non-trivial. We'll show this by showing the induced map on $\pi_n$ is non-trivial.

So, consider the induced map on $\pi_n$. For $n\geq 5$ odd (except $n=7$), Kervaire has shown that $\pi_n(BO(n))\cong \pi_{n-1}(O(n))$ is either $\mathbb{Z}/2\mathbb{Z}$ or $(\mathbb{Z}/2\mathbb{Z})^2$. When $n\neq 7$, the fact that $E$ is non-trivial implies that $\phi_\ast$ is non-trivial, so, for $n\neq 7$, the kernel of the map $\phi_\ast:\pi_n(S^n)\rightarrow \pi_n(BO(n))$ is the even integers. On the other hand, by our choice of $E$ when $n=7$, we have $\ker \phi_\ast = 12\mathbb{Z}$. In either case, the image of $(g\circ f)_\ast:\pi_n(S^n)\rightarrow \pi_n(S^n)$ is multiplication by the odd number $k^r$, so not contained in $\ker \phi_\ast$. The completes the sketch. $\square$.

$\endgroup$
1
  • 2
    $\begingroup$ The only place I use "simply connected" is in the paragraph giving the map $g$. However it's necessary there. For example, if $N$ denotes the quotient of $S^2\times S^3$ by the diagonal action of $\mathbb{Z}/2\mathbb{Z}$ acting as the antipodal map on the first factor and via a degree $-1$ reflection on the second factor, then $N$ is an orientable rational homology sphere, but any map $S^5\rightarrow N$ has degree $0$. Indeed, any such map lifts to $S^2\times S^3$, where the cohomology ring structure easily implies the claim. $\endgroup$ Commented Mar 23, 2022 at 17:29

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .